CN111260136A - Building short-term load prediction method based on ARIMA-LSTM combined model - Google Patents
Building short-term load prediction method based on ARIMA-LSTM combined model Download PDFInfo
- Publication number
- CN111260136A CN111260136A CN202010055212.5A CN202010055212A CN111260136A CN 111260136 A CN111260136 A CN 111260136A CN 202010055212 A CN202010055212 A CN 202010055212A CN 111260136 A CN111260136 A CN 111260136A
- Authority
- CN
- China
- Prior art keywords
- data
- arima
- load
- model
- lstm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Withdrawn
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
Abstract
The invention discloses a building short-term load prediction method based on an ARIMA-LSTM combined model, which comprises the steps of collecting influence factor data through a data collector, and carrying out maximum and minimum normalization processing on the load data and each influence factor data to obtain a dimensionless data set; selecting key influence factors; calculating cosine similarity, and acquiring sample data of similar days as a training set; inputting the load training set of the similar day into an ARIMA-LSTM combined model to obtain a load prediction result; wherein the influencing factor data comprises load data, meteorological data and date type data; when training sample data for building load prediction is analyzed and screened, similar day data sequences are selected by considering the meteorological factors and the grey correlation degree of the date type sequences, and prediction accuracy is effectively improved.
Description
Technical Field
The invention relates to the technical field of space power load prediction, in particular to a building short-term load prediction method based on an ARIMA-LSTM combined model.
Background
With the high-speed development of urban construction in China, the proportion of building energy consumption is continuously increased; in order to relieve the energy crisis and improve the environmental deterioration, reduce the building energy consumption and improve the energy efficiency management, the important attention of the industry is paid, and further the building load prediction becomes an important research content of the ubiquitous power internet of things; the accurate load prediction provides decision basis for the building energy efficiency management system to make electricity demand response and load scheduling planning, is beneficial to optimizing supply and demand balance, improves the utilization rate of electric equipment, and has important significance for energy-saving scheduling and stable operation of the intelligent power grid.
Currently, the research on short-term load prediction models is mainly divided into two categories: one is a linear model, for example, an ARIMAX model is established to predict short-term load by applying a co-integration theory and considering a temperature sequence with input variables, so that the prediction precision is greatly improved; the regARIMA model is also adopted to predict the monthly load data without the influence of outliers, so that the prediction effect is improved; the research objects of the model are all regional loads, the change rule of the model has strong periodicity, and if the model is thinned to a single building load, due to the fact that the volatility and the randomness of sample data are large, prediction errors can be increased by adopting a linear model; the other type is a nonlinear model, a DBN-SVM combined model is adopted to predict the load in the future hour, the prediction precision is high, and the relation among load data of different time sequences is ignored; an Attention mechanism (Attention) is combined with a long-short-term neural network (LSTM), input characteristics playing a key role in load prediction are highlighted, a multi-step-length multivariable short-term load prediction model is established, and prediction accuracy is effectively improved; the above studies all optimize the nonlinear model, and the prediction effect is good, but as the building load of the time series, the modeling effect by using a single artificial neural network is not necessarily the best due to the problems of misjudgment, insufficient fitting or over fitting of the prediction model. Thus, combining linear and non-linear models may improve the accuracy of the system.
Meanwhile, in the process of predicting by utilizing linear and nonlinear combined models, a combined prediction method is adopted to process a time sequence and an error sequence respectively, so that the prediction precision is improved, wherein the error sequence is obtained by the difference between an original sequence and linear prediction; since there is no linear relationship in the error sequence, it is reasonable to use a non-linear model to deal with some possible non-linear relationships in the error sequence.
Disclosure of Invention
This section is for the purpose of summarizing some aspects of embodiments of the invention and to briefly introduce some preferred embodiments. In this section, as well as in the abstract and the title of the invention of this application, simplifications or omissions may be made to avoid obscuring the purpose of the section, the abstract and the title, and such simplifications or omissions are not intended to limit the scope of the invention.
The invention is provided in view of the problems that the existing building short-term load prediction method based on the ARIMA-LSTM combined model has single and inaccurate building short-term load prediction precision.
Therefore, the invention aims to provide a building short-term load prediction method based on an ARIMA-LSTM combined model.
In order to solve the technical problems, the invention provides the following technical scheme: a building short-term load prediction method based on an ARIMA-LSTM combined model comprises the steps of,
acquiring influence factor data through a data acquisition unit, and performing maximum and minimum normalization processing on the load data and each influence factor data to obtain a dimensionless data set;
selecting key influence factors;
calculating cosine similarity, and acquiring sample data of similar days as a training set;
inputting the load training set of the similar day into an ARIMA-LSTM combined model to obtain a load prediction result;
wherein the influencing factor data comprises load data, meteorological data and date type data.
As a preferred scheme of the building short-term load prediction method based on the ARIMA-LSTM combined model, the method comprises the following steps: the dimensionless dataset is:
in the formula, X*Is a normalized value, X is a sample sequence value, XmaxIs the maximum of X, XminIs the minimum value in X.
As a preferred scheme of the building short-term load prediction method based on the ARIMA-LSTM combined model, the method comprises the following steps: and selecting the key influence factors by adopting a grey correlation degree analysis method.
As a preferred scheme of the building short-term load prediction method based on the ARIMA-LSTM combined model, the method comprises the following steps: the grey correlation degree analysis method comprises the following steps:
obtaining the maximum difference Z and the minimum difference Z of the two poles, wherein the formula is as follows:
in the formula, YjFor the daily feature vector to be predicted, XijThe historical day feature vector is obtained;
note βijIs YjAt XijThe correlation coefficient of (c) is expressed by:
in the formula,. DELTA.Zij=|Yj-XijL, |; rho is a resolution coefficient;
calculating the gray associated weight mu of each influence factorjThe formula is as follows:
calculating the grey correlation degree riAnd further selecting key influence factors, wherein the formula is as follows:
as a preferred scheme of the building short-term load prediction method based on the ARIMA-LSTM combined model, the method comprises the following steps: calculating the cosine similarity by adopting a cosine distance;
wherein, the cosine similarity formula is as follows:
in the formula, cosinFpqIs the cosine distance between two load days, Lp、LqRespectively, key influence factor data of certain two days.
As a preferred scheme of the building short-term load prediction method based on the ARIMA-LSTM combined model, the method comprises the following steps: inputting the similar daily load training set into an ARIMA-LSTM combined model to obtain a load prediction result, wherein the step of obtaining the load prediction result comprises the following steps:
inputting the load training set data of the similar day into an ARIMA model to obtain a linear load predicted value;
comparing the load training data with the linear fitting data to obtain a fitting error sequence;
combining key meteorological factor data of similar days, and predicting a fitting error sequence by using an LSTM model;
and adding the linear fitting predicted value and the nonlinear error predicted value to obtain a final load prediction result.
As a preferred scheme of the building short-term load prediction method based on the ARIMA-LSTM combined model, the method comprises the following steps: inputting the load training set data of the similar day into an ARIMA model to obtain a linear load predicted value, wherein the formula is as follows:
in the formula: l is the hysteresis order, ytIs a current value, λiFor the auto-regressive coefficients of order p, βiIs a moving average coefficient corresponding to order q, d is a difference order, epsilontIs a random error term.
As a preferred scheme of the building short-term load prediction method based on the ARIMA-LSTM combined model, the method comprises the following steps: and comparing the load training data with the linear fitting data to obtain a fitting error sequence, wherein the calculation formula is as follows:
ΔAi=Hi-Ai,i=1,2,…,n
in the formula: and delta A is a fitting error sequence, H is an original load data sequence, and A is an ARIMA linear model fitting sequence.
As a preferred scheme of the building short-term load prediction method based on the ARIMA-LSTM combined model, the method comprises the following steps: and (3) predicting a fitting error sequence by using an LSTM model in combination with key meteorological factor data of similar days, wherein the calculation formula is as follows:
ΔLt=f(ΔAt-1,ΔAt-2,…,ΔAt-n)+εt
in the formula: f. oftTo forget the door, itTo the input gate otTo output gate, gtAs an alternative state, ctUpdated cell state for current input, htIs the current predicted or output state; wfWeight matrix for forgetting gate, WiIs a weight matrix of input gates, WoIs a weight matrix of output gates, WgA weight matrix which is an alternative state; bfOffset of forgetting door, biAs an offset of the input gate, boOffset of output gate, bgAn offset for the alternative state; sigma is a sigmoid activation function; tan h is a hyperbolic tangent activation function; is an element in a vectorMultiplication by bit; f (-) is expressed as the nonlinear modeling of LSTM, εtIs a random error.
As a preferred scheme of the building short-term load prediction method based on the ARIMA-LSTM combined model, the method comprises the following steps: and adding the linear fitting predicted value and the nonlinear error predicted value to obtain a final load prediction result, wherein the calculation formula is as follows:
PALt=Pt+ΔLt
in the formula, PALtFor final load prediction, PtAnd (4) predicting the linear load of the ARIMA model.
The invention has the beneficial effects that: when training sample data of building load prediction is analyzed and screened, similar day data sequences are selected by considering the meteorological factors and the grey correlation degree of the date type sequences, so that the prediction precision is effectively improved; meanwhile, by carrying out normalization processing on the sample data, the convergence speed of the model is improved; the building load is predicted by adopting a method of combining linear and nonlinear models, linear components in data are eliminated to the maximum extent by utilizing a linear model ARIMA to fit load data of a time sequence, then an error sequence which cannot be fitted by the linear model is predicted by a nonlinear model LSTM, and a final prediction result is corrected, so that the prediction performance of the linear model and the nonlinear model is fully exerted, and the prediction precision is effectively improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the description of the embodiments will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise. Wherein:
FIG. 1 is a schematic flow chart of the overall steps of the building short-term load prediction method based on the ARIMA-LSTM combined model.
FIG. 2 is a schematic flow chart of steps of inputting a similar daily load training set into an ARIMA-LSTM combined model to obtain a load prediction result according to the ARIMA-LSTM combined model-based building short-term load prediction method.
FIG. 3 is a similar day selection flowchart of the building short-term load prediction method based on the ARIMA-LSTM combined model according to the present invention;
FIG. 4 is a standard LSTM network memory unit according to the building short-term load prediction method based on the ARIMA-LSTM combined model;
FIG. 5 is a flow chart of the ARIMA-LSTM combined model prediction method based on the ARIMA-LSTM combined model building short-term load prediction method;
FIG. 6 is a comparison of the daily average absolute error of the load prediction of the ARIMA model, the LSTM model, the ARIMA-SVM model and the ARIMA-LSTM model according to the short-term building load prediction method based on the ARIMA-LSTM model;
FIG. 7 is a comparison of absolute percentage errors of the ARIMA model, the LSTM model, the ARIMA-SVM model and the ARIMA-LSTM model according to the short-term building load prediction method based on the ARIMA-LSTM model.
Detailed Description
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in detail below.
In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, but the present invention may be practiced in other ways than those specifically described and will be readily apparent to those of ordinary skill in the art without departing from the spirit of the present invention, and therefore the present invention is not limited to the specific embodiments disclosed below.
Furthermore, reference herein to "one embodiment" or "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one implementation of the invention. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are separate or alternative embodiments mutually exclusive of other embodiments.
Furthermore, the present invention is described in detail with reference to the drawings, and in the detailed description of the embodiments of the present invention, the cross-sectional view illustrating the structure of the device is not enlarged partially according to the general scale for convenience of illustration, and the drawings are only exemplary and should not be construed as limiting the scope of the present invention. In addition, the three-dimensional dimensions of length, width and depth should be included in the actual fabrication.
Example 1
Referring to fig. 1 and fig. 2, an overall structure schematic diagram of a building short-term load prediction method based on an ARIMA-LSTM combined model is provided, as shown in fig. 1, the building short-term load prediction method based on the ARIMA-LSTM combined model includes that influence factor data are collected through a data collector, and a dimensionless data set is obtained after maximum and minimum normalization processing is performed on the load data and each influence factor data;
selecting key influence factors;
calculating cosine similarity, and acquiring sample data of similar days as a training set;
inputting the load training set of the similar day into an ARIMA-LSTM combined model to obtain a load prediction result;
wherein the influencing factor data comprises load data, meteorological data and date type data.
Specifically, the main body of the present invention includes S1: acquiring influence factor data through a data acquisition unit, and performing maximum and minimum normalization processing on the load data and each influence factor data to obtain a dimensionless data set; the influence factor data comprises load data, meteorological data and date type data;
it should be noted that the dimensionless data set is:
in the formula, X*Is a normalized value, X is a sample sequence value, XmaxIs the maximum of X, XminIs the minimum value in X.
S2: selecting key influence factors, wherein the key influence factors are selected by adopting a grey correlation degree analysis method;
it should be noted that the gray correlation analysis method comprises the following steps:
s21: obtaining the maximum difference Z and the minimum difference Z of the two poles, wherein the formula is as follows:
in the formula, YjFor the daily feature vector to be predicted, XijThe historical day feature vector is obtained;
s22 note βijIs YjAt XijThe correlation coefficient of (c) is expressed by:
in the formula,. DELTA.Zij=|Yj-XijL, |; rho is a resolution coefficient and is generally in the range of [0, 1%]Taking p as 0.5;
s23: calculating the gray associated weight mu of each influence factorjThe formula is as follows:
s24: calculating the grey correlation degree riAnd further selecting key influence factors, wherein the formula is as follows:
s3: calculating cosine similarity, and acquiring sample data of similar days as a training set; it should be noted that the cosine similarity is calculated by using a cosine distance;
wherein, the cosine similarity formula is as follows:
in the formula, cosinFpqIs the cosine distance between two load days, Lp、LqRespectively the key influence factor data of certain two days
S4: inputting the load training set of the similar day into an ARIMA-LSTM combined model to obtain a load prediction result;
further, the step of inputting the similar daily load training set into the ARIMA-LSTM combined model to obtain the load prediction result comprises:
s41: inputting the load training set data of the similar day into an ARIMA model to obtain a linear load predicted value;
inputting the similar daily load training set data into an ARIMA model to obtain a linear load predicted value, wherein the formula is as follows:
in the formula: l is the hysteresis order, ytIs a current value, λiFor the auto-regressive coefficients of order p, βiIs a moving average coefficient corresponding to order q, d is a difference order, epsilontIs a random error term;
s42: comparing the load training data with the linear fitting data to obtain a fitting error sequence;
it should be noted that, the load training data and the linear fitting data are compared to obtain a fitting error sequence, and the calculation formula is as follows:
ΔAi=Hi-Ai,i=1,2,…,n
in the formula: delta A is a fitting error sequence, H is an original load data sequence, and A is an ARIMA linear model fitting sequence;
s43: combining key meteorological factor data of similar days, and predicting a fitting error sequence by using an LSTM model;
the method is characterized in that a fitting error sequence is predicted by using an LSTM model in combination with key meteorological factor data of similar days, and the calculation formula is as follows:
ΔLt=f(ΔAt-1,ΔAt-2,…,ΔAt-n)+εt
in the formula: f. oftTo forget the door, itTo the input gate otTo output gate, gtAs an alternative state, ctUpdated cell state for current input, htIs the current predicted or output state; wfWeight matrix for forgetting gate, WiIs a weight matrix of input gates, WoIs a weight matrix of output gates, WgA weight matrix which is an alternative state; bfOffset of forgetting door, biAs an offset of the input gate, boOffset of output gate, bgAn offset for the alternative state; sigma is a sigmoid activation function; tan h is a hyperbolic tangent activation function; is a bit-wise multiplication of elements in the vector; f (-) is expressed as the nonlinear modeling of LSTM, εtIs a random error;
s44: and adding the linear fitting predicted value and the nonlinear error predicted value to obtain a final load prediction result.
It should be noted that the linear fitting prediction value and the nonlinear error prediction value are added to obtain a final load prediction result, and the calculation formula is as follows:
PALt=Pt+ΔLt
in the formula, PALtFor final load prediction, PtLinear load prediction for the ARIMA model.
Example 2
The embodiment is to verify and explain the technical effects adopted in the method, and the different methods selected in the embodiment and the method adopted in the embodiment are compared and tested, and the test results are compared by means of scientific demonstration to verify the real effect of the method.
Firstly, acquiring building load influence factor data by a data acquisition unit, acquiring data including load data, meteorological data and date type data, wherein the sampling interval of the data is 1 hour, and recording 24 data points every day; and then analyzing the load influence factors when analyzing and screening training sample data of building load prediction, preprocessing the original sequence data, and finally selecting similar day sequence data according to the grey correlation degree and the cosine distance.
As shown in fig. 1, the steps of analyzing the collected data and operating the data are described as follows:
1) the weather factors have an important influence on the building short-term load prediction, and the main influence factors are as follows: air temperature, wind speed, relative humidity, rainfall, etc.; the date type is also another important factor for building loads, such as the amount of electric power load on weekdays, non-weekdays (saturday, sunday, and holidays) is also significantly different.
2) Selecting day characteristic related factors: date type d, maximum temperature tmaxMinimum air temperature tminAnd the wind speed w, the relative humidity h and the rainfall r are used as samples for calculating the grey correlation degree, the sequence data are subjected to fuzzification clustering mapping processing, and a daily characteristic quantity mapping table is shown as the following table:
3) adopting maximum and minimum normalization processing to convert the sample data of each influence factor into a [0,1] interval, wherein the calculation formula is as follows:
in the formula, X*Is a normalized value, X is a sample sequence value, XmaxIs the maximum of X, XminIs the minimum value in X.
4) And selecting key influence factors by adopting a grey correlation degree analysis method.
Let the feature vector of the day to be predicted be Yj=(y1,y2,…,ym) J is 1,2, …, m, and the feature vector of the ith history day is Xij=(xi1,xi2,…,xim) And i is 1,2, …, n, wherein m is the number of influence factors, and n is the number of historical days. By the daily feature vector Y to be predictedjFor reference sequence, historical day feature vector XijIs a comparative sequenceColumns;
the grey correlation degree is calculated by the following steps:
calculating a grey correlation coefficient, firstly, solving a maximum difference and a minimum difference of two poles, wherein the calculation formulas are respectively as follows:
then, βijIs YjAt XijThe correlation coefficient of (b), then:
wherein Δ Zij=|Yj-XijL, |; rho is a resolution coefficient and is generally in the range of [0, 1%]The value of p is usually 0.5.
43) Calculating the gray associated weight of each influence factor:
44) calculating the grey correlation degree:
the grey correlation results for each influencing factor are given in the following table:
serial number | Name of influencing factor | Data range | Degree of |
1 | Air temperature | [0,1] | 0.6031 |
2 | Type of date | [0,1] | 0.1985 |
3 | Amount of rainfall | [0,1] | 0.0676 |
4 | Wind speed | [0,1] | 0.0614 |
5 | Relative humidity | [0,1] | 0.0379 |
5) After the key influence factors are selected, selecting the acquaintance day of building load prediction by taking the factor data sequence as an object; the invention selects cosine distance to carry out similarity measurement, selects 15 historical days which are closest to the date type and the air temperature condition of a prediction day as similar days, and takes the selected 15 similar date type data, air temperature data and load data as a training set of a combined prediction model, and the specific operation is as follows:
setting a data sequence of key influence factors of certain two days as follows: l isp=[lp1,lp2,…,lpt];Lq=[lq1,lq2,…,lqt ]Then the cosine similarity is calculated as follows:
in the formula, cosinFpqCosine distance between two load days, cosinFpqThe larger the value, the smaller the difference of the influence factor sequence curve forms, and the higher the similarity.
Similar day selection results are shown in the following table:
date | Cosine distance | Date | Cosine distance | Date | Cosine distance |
2017/03/08 | 1.9024 | 2017/03/05 | 1.7357 | 2017/02/29 | 1.6025 |
2017/02/28 | 1.8852 | 2017/03/01 | 1.7086 | 2017/02/23 | 1.5846 |
2017/03/02 | 1.8641 | 2017/02/20 | 1.6804 | 2017/03/07 | 1.5549 |
2017/03/09 | 1.8294 | 2017/03/06 | 1.6536 | 2017/02/24 | 1.5308 |
2017/02/16 | 1.7562 | 2017/03/12 | 1.6231 | 2017/02/22 | 1.5128 |
Further, a linear ARIMA model is constructed, and the specific steps are as follows:
the ARIMA (p, d, q) model is composed of an autoregressive model AR (p) of order p, a moving average model MA (q) of order q and the difference times d for generating a stable sequence, the non-stable time sequence is stabilized through difference processing, and then the hysteresis value of a dependent variable and the current value and the hysteresis value of a random error term are regressed. The optimal order of the model is determined according to the Chichi information criterion (AIC), and the ARIMA (2,1,2) model is finally established, so that the requirement on model stability is met, and random fluctuation existing in prediction is effectively eliminated.
Based on the principle of FIG. 4, a nonlinear LSTM model is analyzed and constructed, and the specific principle and steps are as follows:
neural networks are a very efficient tool for approximating any non-linear function and are widely used in dealing with non-linear problems. The recurrent neural network is mainly used for processing and predicting time sequence data and is trained through a time back propagation or real-time recurrent learning algorithm.
In practice, the problem of gradient disappearance or explosion during optimization can be caused by overlong sequence, so that the training result is often poor; the invention employs a long-short term memory (LSTM) neural network, as a variant of RNN, specifically designed to overcome the gradient vanishing problem, enabling long-term storage of information.
The LSTM network is composed of an input layer, an output layer, and a plurality of hidden layers interposed therebetween. Wherein the hidden layer is constructed into units with memory function, each unit comprises three gates, namely an Input Gate (Input Gate), a forgetting Gate (Forgetgate), and an Output Gate (Output Gate); standard LSTM network memory cell As shown in FIG. 2, in the forgetting gate, each input xtPrevious time unit output ht-1Previous time cell state ct-1A forgetting part for jointly determining the state memory cells; in the input gate, the unit state c is activated under the activation of sigmoid and tanh functionst-1Update to ct(ii) a Updated cell state c in output gatetH is selectively output through sigmoid and tanh functions againt. The LSTM network element may be defined by the following system of equations:
ΔLt=f(ΔAt-1,ΔAt-2,…,ΔAt-n)+εt
in the formula: f. oftTo forget the door, itTo the input gate otTo output gate, gtAs an alternative state, ctUpdated cell state for current input, htIs the current predicted or output state;Wfweight matrix for forgetting gate, WiIs a weight matrix of input gates, WoIs a weight matrix of output gates, WgA weight matrix which is an alternative state; bfOffset of forgetting door, biAs an offset of the input gate, boOffset of output gate, bgAn offset for the alternative state; sigma is a sigmoid activation function; tan h is a hyperbolic tangent activation function; is a bit-wise multiplication of elements in the vector; f (-) is expressed as the nonlinear modeling of LSTM, εtIs a random error.
The hidden layer of the LSTM network is set to have 50 neurons, the training block size (batch _ size) is set to 32, and the training times (epochs) is set to 50.
The specific operation steps based on fig. 5 are described as follows:
1) let the original load data set be H ═ H1,h2,…,hn]Obtaining a historical data fitting sequence A ═ A of H by using an ARIMA linear model1,A2,…,An]And a prediction sequence PA=[P1,P2,…,Pt]And comparing the fitting value with the actual load value to obtain a fitting error sequence delta A ═ delta A1,ΔA2,…,ΔAn]The calculation formula is as follows:
ΔAi=Hi-Ai,i=1,2,…,n
2) the fitting error sequence delta A reduces the influence of linear components to a certain extent, and the nonlinear characteristic is strong. The nonlinear learning performance of the neural network model is suitable for the prediction and correction of the fitting error sequence, so that the fitting error sequence delta A is predicted by using a long-short term memory neural network (LSTM) to obtain a nonlinear error prediction term delta L, and the calculation formula is as follows:
ΔLt=f(ΔAt-1,ΔAt-2,…,ΔAt-n)+εt
in the formula: f (-) is expressed as the nonlinear modeling of LSTM, εtRandom errors, unpredictable.
3) Will predict the value Δ LtFurther learning the non-linearity of the error sequence as a result of the load error predictionThe method is characterized in that a stable and effective prediction result is output and error correction is carried out, and the prediction result of the ARIMA-LSTM combined model is finally obtained, wherein the calculation formula is as follows:
PALt=Pt+ΔLt
and evaluating the load prediction performance of the model by selecting statistical indexes such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE) and the like. The calculation formula is as follows:
in the formula, N is the total number of the predicted points,as the actual value of the load, PtThe load is predicted value.
In order to compare the prediction effects of different models, the invention adopts an ARIMA model, an LSTM neural network model, an ARIMA-SVM combined model and an ARIMA-LSTM combined model to predict the load of 24 moments (time interval 1h) each day from 3, 13 to 19 days of 2012 of an office building respectively, and calculates the average absolute error e of each dayMAPEThe prediction result error curves of the models are shown in FIG. 6, the average absolute error of the load of the method in 7 consecutive days of prediction is smaller than that of the other three methods, and the prediction effect is the best.
In order to further analyze the performance of the model, the invention respectively predicts the load of a certain day in 3 months of 2012 by using an ARIMA model, an LSTM model, an ARIMA-SVM combined model and an ARIMA-LSTM combined model, and calculates the absolute error percentage of the load prediction at each moment by comparing the predicted value of each load model with the actual load data, as shown in FIG. 7.
Respectively counting the probability that the prediction error at each moment is less than 3% according to the error percentage distribution of the load prediction of each model in the graph 7; the probability that the prediction error of the method is less than 3% is far higher than that of other three models, so that the method has higher accuracy in load prediction; at the same time, the bookMean absolute error e of ARIMA-LSTM combined model used hereinMAPE2.02%, which is lower than 3.13%, 1.42% and 0.8% respectively compared with ARMIA model, LSTM model and ARIMA-SVM model, and has average absolute error eMAERoot mean square error eRMSEAnd the method is also smaller than other models, so the method has more ideal effect when controlling the prediction error.
It should be appreciated that in the development of any such actual implementation, as in any engineering or design project, numerous implementation-specific decisions may be made. Such a development effort might be complex and time consuming, but would nevertheless be a routine undertaking of design, fabrication, and manufacture for those of ordinary skill having the benefit of this disclosure, without undue experimentation.
It should be noted that the above-mentioned embodiments are only for illustrating the technical solutions of the present invention and not for limiting, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions may be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, which should be covered by the claims of the present invention.
Claims (10)
1. A building short-term load prediction method based on an ARIMA-LSTM combined model is characterized by comprising the following steps: comprises the steps of (a) carrying out,
acquiring influence factor data through a data acquisition unit, and performing maximum and minimum normalization processing on the load data and each influence factor data to obtain a dimensionless data set;
selecting key influence factors;
calculating cosine similarity, and acquiring sample data of similar days as a training set;
inputting the load training set of the similar day into an ARIMA-LSTM combined model to obtain a load prediction result;
wherein the influencing factor data comprises load data, meteorological data and date type data.
3. The building short-term load prediction method based on ARIMA-LSTM combined model as claimed in claim 1 or 2, characterized in that: and selecting the key influence factors by adopting a grey correlation degree analysis method.
4. The building short-term load prediction method based on the ARIMA-LSTM combined model as recited in claim 3, wherein: the grey correlation degree analysis method comprises the following steps:
obtaining the maximum difference Z and the minimum difference Z of the two poles, wherein the formula is as follows:
in the formula, YjFor the daily feature vector to be predicted, XijThe historical day feature vector is obtained;
note βijIs YjAt XijThe correlation coefficient of (c) is expressed by:
in the formula, △ Zij=|Yj-XijL, |; rho is a resolution coefficient;
calculating the gray associated weight mu of each influence factorjThe formula is as follows:
calculating the grey correlation degree riAnd further selecting key influence factors, wherein the formula is as follows:
5. the building short-term load prediction method based on the ARIMA-LSTM combined model as recited in claim 4, wherein: the cosine similarity is calculated by adopting a cosine distance;
wherein, the cosine similarity formula is as follows:
in the formula, cosinFpqIs the cosine distance between two load days, Lp、LqRespectively, key influence factor data of certain two days.
6. The building short-term load prediction method based on the ARIMA-LSTM combined model as claimed in claim 4 or 5, characterized in that: the step of inputting the similar daily load training set into the ARIMA-LSTM combined model to obtain the load prediction result comprises the following steps:
inputting the load training set data of the similar day into an ARIMA model to obtain a linear load predicted value;
comparing the load training data with the linear fitting data to obtain a fitting error sequence;
combining key meteorological factor data of similar days, and predicting a fitting error sequence by using an LSTM model;
and adding the linear fitting predicted value and the nonlinear error predicted value to obtain a final load prediction result.
7. The building short-term load prediction method based on the ARIMA-LSTM combined model as recited in claim 6, wherein: inputting the load training set data of the similar day into an ARIMA model to obtain a linear load predicted value, wherein the formula is as follows:
in the formula: l is the hysteresis order, ytIs a current value, λiFor the auto-regressive coefficients of order p, βiIs a moving average coefficient corresponding to order q, d is a difference order, epsilontIs a random error term.
8. The building short-term load prediction method based on the ARIMA-LSTM combined model as recited in claim 7, wherein: and comparing the load training data with the linear fitting data to obtain a fitting error sequence, wherein the calculation formula is as follows:
△Ai=Hi-Ai,i=1,2,…,n
wherein △ A is the fitting error sequence, H is the original load data sequence, and A is the ARIMA linear model fitting sequence.
9. The building short-term load prediction method based on the ARIMA-LSTM combined model as recited in claim 8, wherein: the method is characterized in that the LSTM model is used for predicting the fitting error sequence by combining the key meteorological factor data of similar days, and the calculation formula is as follows:
△Lt=f(△At-1,△At-2,…,△At-n)+εt
in the formula: f. oftTo forget the door, itTo the input gate otTo output gate, gtAs an alternative state, ctUpdated cell state for current input, htIs the current predicted or output state; wfWeight matrix for forgetting gate, WiIs a weight matrix of input gates, WoIs a weight matrix of output gates, WgA weight matrix which is an alternative state; bfOffset of forgetting door, biAs an offset of the input gate, boOffset of output gate, bgAn offset for the alternative state; sigma is a sigmoid activation function; tan h is a hyperbolic tangent activation function; is a bit-wise multiplication of elements in the vector; f (-) is expressed as the nonlinear modeling of LSTM, εtIs a random error.
10. A building short term load prediction method based on ARIMA-LSTM combined model as claimed in claim 8 or 9, characterized by: and adding the linear fitting predicted value and the nonlinear error predicted value to obtain a final load prediction result, wherein the calculation formula is as follows:
PALt=Pt+△Lt
in the formula, PALtFor final load prediction, PtAnd (4) predicting the linear load of the ARIMA model.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010055212.5A CN111260136A (en) | 2020-01-17 | 2020-01-17 | Building short-term load prediction method based on ARIMA-LSTM combined model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010055212.5A CN111260136A (en) | 2020-01-17 | 2020-01-17 | Building short-term load prediction method based on ARIMA-LSTM combined model |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111260136A true CN111260136A (en) | 2020-06-09 |
Family
ID=70954399
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010055212.5A Withdrawn CN111260136A (en) | 2020-01-17 | 2020-01-17 | Building short-term load prediction method based on ARIMA-LSTM combined model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111260136A (en) |
Cited By (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111832809A (en) * | 2020-06-19 | 2020-10-27 | 山东大学 | Building energy consumption load prediction method and system based on Holt-Winters and extreme learning machine |
CN111932402A (en) * | 2020-08-13 | 2020-11-13 | 河南理工大学 | Short-term power load bidirectional combined prediction method based on similar day and LSTM |
CN111931992A (en) * | 2020-07-16 | 2020-11-13 | 广东电网有限责任公司 | Power load prediction index selection method and device |
CN111984503A (en) * | 2020-08-17 | 2020-11-24 | 网宿科技股份有限公司 | Method and device for identifying abnormal data of monitoring index data |
CN112163702A (en) * | 2020-09-24 | 2021-01-01 | 国网江苏省电力有限公司徐州供电分公司 | Prediction method and device for stable section load rate, electronic device and storage medium |
CN112232479A (en) * | 2020-09-11 | 2021-01-15 | 湖北大学 | Building energy consumption space-time factor characterization method based on deep cascade neural network and related products |
CN112418485A (en) * | 2020-10-27 | 2021-02-26 | 西安交通大学 | Household load prediction method and system based on load characteristics and power consumption behavior mode |
CN112434787A (en) * | 2020-10-28 | 2021-03-02 | 西安交通大学 | Terminal space energy consumption prediction method based on building total energy consumption, medium and equipment |
CN112488404A (en) * | 2020-12-07 | 2021-03-12 | 广西电网有限责任公司电力科学研究院 | Multithreading efficient prediction method and system for large-scale power load of power distribution network |
CN112508286A (en) * | 2020-12-11 | 2021-03-16 | 上海电力大学 | Short-term load prediction method based on Kmeans-BilSTM-DMD model |
CN112561728A (en) * | 2020-10-28 | 2021-03-26 | 西安交通大学 | Attention mechanism LSTM-based comprehensive energy consumption cost optimization method, medium and equipment |
CN112734075A (en) * | 2020-11-27 | 2021-04-30 | 北京国电通网络技术有限公司 | Building cold/heat/electric load prediction model and algorithm system |
CN112736902A (en) * | 2020-12-24 | 2021-04-30 | 国网山西省电力公司 | STL decomposition-based time series short-term power load prediction method |
CN112966851A (en) * | 2021-01-28 | 2021-06-15 | 国网浙江省电力有限公司金华供电公司 | Short-term load change trend prediction method |
CN112990587A (en) * | 2021-03-24 | 2021-06-18 | 北京市腾河智慧能源科技有限公司 | Method, system, equipment and medium for accurately predicting power consumption of transformer area |
CN113324648A (en) * | 2021-07-09 | 2021-08-31 | 中国铁道科学研究院集团有限公司铁道科学技术研究发展中心 | Portable high-speed railway wheel rail vibration space-time synchronization test method and system |
CN113393057A (en) * | 2021-07-13 | 2021-09-14 | 四川农业大学 | Wheat yield integrated prediction method based on deep fusion machine learning model |
CN113571138A (en) * | 2021-08-10 | 2021-10-29 | 郑州大学 | Alkali-activated mortar bonding performance analysis method based on gray correlation and weight contribution |
CN114399024A (en) * | 2021-12-20 | 2022-04-26 | 淮阴工学院 | Oil gas concentration big data intelligent detection system |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104978611A (en) * | 2015-07-06 | 2015-10-14 | 东南大学 | Neural network photovoltaic power generation output prediction method based on grey correlation analysis |
CN105760947A (en) * | 2016-02-02 | 2016-07-13 | 国网江苏省电力公司 | Method for analyzing influence factors of meteorological environment on network load |
CN109583645A (en) * | 2018-11-28 | 2019-04-05 | 上海电力学院 | A kind of public building short-term load forecasting method |
CN110689195A (en) * | 2019-09-26 | 2020-01-14 | 云南电网有限责任公司电力科学研究院 | Power daily load prediction method |
-
2020
- 2020-01-17 CN CN202010055212.5A patent/CN111260136A/en not_active Withdrawn
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104978611A (en) * | 2015-07-06 | 2015-10-14 | 东南大学 | Neural network photovoltaic power generation output prediction method based on grey correlation analysis |
CN105760947A (en) * | 2016-02-02 | 2016-07-13 | 国网江苏省电力公司 | Method for analyzing influence factors of meteorological environment on network load |
CN109583645A (en) * | 2018-11-28 | 2019-04-05 | 上海电力学院 | A kind of public building short-term load forecasting method |
CN110689195A (en) * | 2019-09-26 | 2020-01-14 | 云南电网有限责任公司电力科学研究院 | Power daily load prediction method |
Non-Patent Citations (3)
Title |
---|
JAVIER CONTRERAS ET AL.: "ARIMA Models to Predict Next-Day Electricity Prices", 《POWER ENGINEERING REVIEW, IEEE》 * |
李鹏辉 等: "基于ARIMA-LSTM组合模型的楼宇短期负荷预测方法研究", 《上海电力学院学报》 * |
邹平 著: "《金融学系列 "十二五"普通高等教育本科国家规划教材 金融计量学 第4版》", 31 August 2018, 上海财经大学出版社 * |
Cited By (28)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111832809A (en) * | 2020-06-19 | 2020-10-27 | 山东大学 | Building energy consumption load prediction method and system based on Holt-Winters and extreme learning machine |
CN111931992A (en) * | 2020-07-16 | 2020-11-13 | 广东电网有限责任公司 | Power load prediction index selection method and device |
CN111932402B (en) * | 2020-08-13 | 2023-05-30 | 河南理工大学 | Short-term power load bidirectional combination prediction method based on similar day and LSTM |
CN111932402A (en) * | 2020-08-13 | 2020-11-13 | 河南理工大学 | Short-term power load bidirectional combined prediction method based on similar day and LSTM |
CN111984503A (en) * | 2020-08-17 | 2020-11-24 | 网宿科技股份有限公司 | Method and device for identifying abnormal data of monitoring index data |
CN111984503B (en) * | 2020-08-17 | 2023-12-01 | 网宿科技股份有限公司 | Method and device for identifying abnormal data of monitoring index data |
CN112232479A (en) * | 2020-09-11 | 2021-01-15 | 湖北大学 | Building energy consumption space-time factor characterization method based on deep cascade neural network and related products |
CN112163702A (en) * | 2020-09-24 | 2021-01-01 | 国网江苏省电力有限公司徐州供电分公司 | Prediction method and device for stable section load rate, electronic device and storage medium |
CN112163702B (en) * | 2020-09-24 | 2023-08-08 | 国网江苏省电力有限公司徐州供电分公司 | Method and device for predicting stable section load rate, electronic equipment and storage medium |
CN112418485A (en) * | 2020-10-27 | 2021-02-26 | 西安交通大学 | Household load prediction method and system based on load characteristics and power consumption behavior mode |
CN112434787A (en) * | 2020-10-28 | 2021-03-02 | 西安交通大学 | Terminal space energy consumption prediction method based on building total energy consumption, medium and equipment |
CN112561728B (en) * | 2020-10-28 | 2023-07-11 | 西安交通大学 | Comprehensive energy consumption optimization method, medium and device based on attention mechanism LSTM |
CN112434787B (en) * | 2020-10-28 | 2024-04-05 | 西安交通大学 | Terminal space energy consumption prediction method, medium and equipment based on total energy consumption of building |
CN112561728A (en) * | 2020-10-28 | 2021-03-26 | 西安交通大学 | Attention mechanism LSTM-based comprehensive energy consumption cost optimization method, medium and equipment |
CN112734075A (en) * | 2020-11-27 | 2021-04-30 | 北京国电通网络技术有限公司 | Building cold/heat/electric load prediction model and algorithm system |
CN112488404A (en) * | 2020-12-07 | 2021-03-12 | 广西电网有限责任公司电力科学研究院 | Multithreading efficient prediction method and system for large-scale power load of power distribution network |
CN112508286A (en) * | 2020-12-11 | 2021-03-16 | 上海电力大学 | Short-term load prediction method based on Kmeans-BilSTM-DMD model |
CN112736902A (en) * | 2020-12-24 | 2021-04-30 | 国网山西省电力公司 | STL decomposition-based time series short-term power load prediction method |
CN112736902B (en) * | 2020-12-24 | 2023-04-04 | 国网山西省电力公司 | STL decomposition-based time series short-term power load prediction method |
CN112966851A (en) * | 2021-01-28 | 2021-06-15 | 国网浙江省电力有限公司金华供电公司 | Short-term load change trend prediction method |
CN112990587A (en) * | 2021-03-24 | 2021-06-18 | 北京市腾河智慧能源科技有限公司 | Method, system, equipment and medium for accurately predicting power consumption of transformer area |
CN112990587B (en) * | 2021-03-24 | 2023-10-24 | 北京市腾河智慧能源科技有限公司 | Method, system, equipment and medium for accurately predicting power consumption of transformer area |
CN113324648A (en) * | 2021-07-09 | 2021-08-31 | 中国铁道科学研究院集团有限公司铁道科学技术研究发展中心 | Portable high-speed railway wheel rail vibration space-time synchronization test method and system |
CN113393057A (en) * | 2021-07-13 | 2021-09-14 | 四川农业大学 | Wheat yield integrated prediction method based on deep fusion machine learning model |
CN113571138A (en) * | 2021-08-10 | 2021-10-29 | 郑州大学 | Alkali-activated mortar bonding performance analysis method based on gray correlation and weight contribution |
CN113571138B (en) * | 2021-08-10 | 2024-03-29 | 郑州大学 | Grey correlation and weight contribution-based alkali-activated mortar bonding performance analysis method |
CN114399024B (en) * | 2021-12-20 | 2023-02-03 | 淮阴工学院 | Oil gas concentration big data intelligent detection system |
CN114399024A (en) * | 2021-12-20 | 2022-04-26 | 淮阴工学院 | Oil gas concentration big data intelligent detection system |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111260136A (en) | Building short-term load prediction method based on ARIMA-LSTM combined model | |
CN113962364B (en) | Multi-factor power load prediction method based on deep learning | |
CN112949945B (en) | Wind power ultra-short-term prediction method for improving bidirectional long-term and short-term memory network | |
CN110705743B (en) | New energy consumption electric quantity prediction method based on long-term and short-term memory neural network | |
CN112990556A (en) | User power consumption prediction method based on Prophet-LSTM model | |
CN112149879B (en) | New energy medium-and-long-term electric quantity prediction method considering macroscopic volatility classification | |
CN109242265B (en) | Urban water demand combined prediction method based on least square sum of errors | |
CN113554466B (en) | Short-term electricity consumption prediction model construction method, prediction method and device | |
CN112329990A (en) | User power load prediction method based on LSTM-BP neural network | |
CN111915092A (en) | Ultra-short-term wind power prediction method based on long-time and short-time memory neural network | |
CN113705877A (en) | Real-time monthly runoff forecasting method based on deep learning model | |
CN115169703A (en) | Short-term power load prediction method based on long-term and short-term memory network combination | |
CN115860177A (en) | Photovoltaic power generation power prediction method based on combined machine learning model and application thereof | |
CN115545333A (en) | Method for predicting load curve of multi-load daily-type power distribution network | |
CN115759415A (en) | Power consumption demand prediction method based on LSTM-SVR | |
CN112990587A (en) | Method, system, equipment and medium for accurately predicting power consumption of transformer area | |
CN114117852B (en) | Regional heat load rolling prediction method based on finite difference working domain division | |
CN114595861A (en) | MSTL (modeling, transformation, simulation and maintenance) and LSTM (least Square TM) model-based medium-and-long-term power load prediction method | |
Huang et al. | A decomposition‐based multi‐time dimension long short‐term memory model for short‐term electric load forecasting | |
CN112288187A (en) | Big data-based electricity sales amount prediction method | |
CN116703644A (en) | Attention-RNN-based short-term power load prediction method | |
CN114819395A (en) | Industry medium and long term load prediction method based on long and short term memory neural network and support vector regression combination model | |
CN115293406A (en) | Photovoltaic power generation power prediction method based on Catboost and Radam-LSTM | |
CN112529268A (en) | Medium-short term load prediction method and device based on manifold learning | |
Sun et al. | Information entropy based neural network model for short-term load forecasting |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WW01 | Invention patent application withdrawn after publication | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20200609 |